Chip cooling breakthrough will reduce data center power costs
Traditional passive heatsinks affixed to microprocessors for cooling don’t work well enough for today’s high-speed computations and data throughputs and should be junked, says a group of mechanical engineering researchers.
A better option, they say, are “spirals or mazes that coolant can travel through” within tiny channels on the actual processor. That technique could massively improve efficiency, says Scott Schiffres, an assistant professor at Binghamton University in New York, in an article on the school’s website. The school has developed this new method for cooling chips.
Schiffres and graduate students Arad Azizi and Matthias A. Daeumer who worked on the study say the technique will keep electronics cooler by 18 degrees F and that power use in data centers could be reduced by 5 percent.
they say, bonds a microchannel, 3D-printing-like, additive-printed alloy onto chip-silicon during manufacture instead of using the traditional method of sticking on a heatsink.
Currently, heatsinks, often made up of multiple copper or aluminum fins and affixed with a thermal paste, conduct heat away from the chip. They’re able to do so partly because they have a larger surface than the chip surface and because they use heat-conducting materials such as aluminum. The chips can then run faster without overheating and failing. The heat is usually dissipated into the surrounding air or into water.
“For the heatsink to work, it has to be attached to the CPU, or the graphics processor via a thermal interface material, such as thermal paste,” the university explains.
The problem is that method is inherently inefficient. The adhesive, thermal interface material, while importantly filling the microscopic gaps between the heatsink and the chip (and also stopping the heat sink from falling off), isn’t as good as something completely seamless. Up to now, that has been impossible to achieve—the heat sink wouldn’t stick, for one thing, and gaps would be introduced, thus interrupting the heat pass-through.
Printing cooling microchannels on the chip
The Binghamton researchers say their additive printed technique solves this issue by robustly bonding the cooling mechanism directly to the silicon, bypassing any interface. “We plan to print microchannels on the chip itself,” Schiffres says.
They’re using a tin-silver-titanium alloy that’s 1,000 times thinner than a human hair to perform the metal bond. A melting laser prints the heat-dissipating channels directly onto the silicon in a sub-millisecond operation. The microprocessors thus bypass the need for the typical two layers of thermal paste materials and what’s called the “lid” — a heat spreader layer between the heatsink and chip.
It’s not been easy, they say. Metals and alloys, on the whole, don’t stick to silicon well — strength is compromised, the researchers explain in their paper published in Additive Manufacturing (pdf). There are also issues with thermal expansion mismatches.
Reducing electricity costs and saving the planet
Schiffres says, though, that the invention works and will not only make electronics and data centers more efficient — saving data centers $438 million dollars in electricity annually, he says — but it will also help the planet by stopping 3.7 billion pounds of carbon dioxide from being emitted due to electricity production.
“It will also reduce toxic electronic waste by about 10 million metric tons, enough to fill 25 Empire State Buildings,” Schiffres claims. That’s due to “lower rates of heat-based device failure.”
Fast-overheating graphic processors will, in particular, benefit. In fact, it was computer game players who gave the team their idea — gamers often remove the heatsink lid, along with one of the layers of paste, on graphics cards to improve heat pass-through.
“It will mean big changes for high-end electronics, data centers, and computationally intense programs such as video editing tools and video games,” Schiffres says.